)Effective ASIFT Wide-baseline Stereo Image Matching Based onAdaptive Relative Affine Transformation Stimulation

Objective Robust affine-SIFT(ASIFT)algorithm is attractive to wide-baseline stereo image match ap-plications.However,due to exhaustive image stimulation covering a whole 3D viewpoint space,ASIFT is quite prohibitive in practice due to its huge computation demands from image feature de-scription extraction and high-dimension vector distance calculation.By adaptively selecting the mostvalid fewer affine transformations and implementing less image stimulation,ASIFT is re-developed toachieve high efficiency suitable for the needs of real applications.First,the theory behind ASIFT forstimulating single images based on absolute viewpoint direction is outlined.Also,a relative affinetransformation stimulation model for stereo images is formulated and estimated with homographymapping,and photogrammetric computation.Afterwards,we propose a single progressive computa-tion procedure for quickly obtaining fewer conjunctive points and the parameters for relative affine transfor-mation stimulation.Steps to select valid affine transformations are also summarized.Next,SIFT is imple-mented to select fewer stimulated images to obtain reliable match results.Finally,a sequence of wide-baselinestereo images is tested with proposed approaches and some conclusions are drawn.